Economic Performance, Management Competence and
the Outcome of the Next General Election
David SANDERS
University of Essex
Working Paper n.124
Barcelona 1996
A number of models of UK electoral preferences, based on aggregate-level
popularity functions, have been developed over the last 25 years or so1. The
early models all assumed that there were direct connections between the
condition of the macroeconomy (generally measured in terms of unemployment
and inflation) and the pattern of support (as indicated by monthly or quarterly
opinion poll ratings) for the major political parties: governments were «rewarded»
with greater support if their performance was «good»; they were «punished» by
reduced support if performance was «poor»2. The main aim of these analyses
was typically to establish how much extra support a government could expect to
obtain if it reduced either unemployment or inflation (or, even better, both) by,
say, a percentage point over a 3-month period. During the 1980s analysts began
to include voters’ perceptions of the state of the economy, rather than just its
«objective» condition, as predictors of political support3. These analyses
suggested that voters’ perceptions of what is happening to the economy appear
to be more important in determining patterns of political support than the
economy’s objective condition -though, obviously, there are important
connections between macroeconomic changes and electors’ economic
perceptions. In the l990s, analysts have finally begun to follow through the
statistical logic of their popularity functions. They have ceased to employ these
functions simply as devices for analysing the past fortunes of political parties and
have started to explore their utility for forecasting future movements in political
support4.
I have reviewed elsewhere the success of one simple forecasting model
that correctly predicted the outcome of the 1992 British general election -some 18
months in advance of the event5. In this paper I discuss the prospects for
successfully forecasting the outcome of the next UK general election. The paper’s
main methodological concern is with the problem of deciding which estimation
period should be used in order to produce a particular set of political forecasts.
The principle substantive aim to develop a series of models that simulate the
likely political fortunes of the present government over the next year or so. Given
that Labour popularity generally varies inversely with that of the Conservatives6, I
concentrate solely on developing a suitable popularity function for the
Conservative Party.
The first part of the paper briefly reviews some of the evidence which
supports the conclusion that UK voters’ subjective economic perceptions are
more important than objective economic realities in determining their political
preferences. The second part develops a series of forecasting equations based
partly on different estimation periods and partly on different assumptions about
which particular economic perceptions will prove decisive in British politics in the
run-up to the next general election. Although the precise results of the various
forecasting models differ, they all converge on the conclusion that the
Conservatives cannot recover sufficiently on the basis of purely economic
considerations to win that election.
THE IMPORTANCE OF ECONOMIC PERCEPTIONS
Figures 1-7 show the general connections (or lack of them) between
Conservative Party popularity and a range of measures of economic
performance. (I assume throughout the ensuing discussion that increases in
unemployment, inflation, interest rates and taxation7 connote «bad» performance
whereas increases in disposable income connote «good» performance). With the
possible exception of the two interest rate variables and taxation (which tend to
trend upwards when popularity trends downwards and vice versa) what is
remarkable about the various figures is the lack of any obvious visual relationship
between popularity and economic conditions. Indeed, as Table Al in the Appendix
indicates, this conclusion is also supported by more formal statistical models.
These models show that, when appropriate controls are made for aggregate
economic perceptions and for certain key political events, unemployment,
inflation, personal disposable income and interest rates -whether measured as
levels, changes in levels or rates of change; whether lagged or unlagged- all fail
to exert direct effects on Conservative popularity.
To analysts unfamiliar with recent work on British popularity functions, this
conclusion may seem surprising. What has become increasingly clear over the
last decade or so, however, is that the effects of the real economy on UK voters’
political preferences are strongly mediated by voters’ economic perceptions. The
relationship between incumbent party popularity and one of the pivotal sets of
economic perceptions is shown in Figure 88. As the figure shows, when
aggregate personal economic expectations rise, the Conservatives’ poll ratings
also tend to rise -and vice versa. The theoretical interpretation of this relationship
is simple. If I am optimistic about my economic prospects, I am more inclined to
seek to preserve the status quo that has produced my optimism; if I am
pessimistic about my economic prospects, I am more inclined to seek to change
the status quo that has produced my pessimism. As Figure 8 clearly
demonstrates, the general elections of 1983, 1987 and 1992 were all preceded by
a lengthy period of gradually rising personal expectations -the ideal platform for
an incumbent government to secure its own re-election.
More formal statistical models of the relationship between personal
expectations and government popularity show that it is both stable and robust
over the 1979-1994 period9. The balance of evidence also suggests that it is
personal expectations that have exerted a causal influence upon Conservative
support rather than the other way round. Simple exogeneity tests confirm that,
whereas lagged values of personal expectations predict popularity, lagged values
of popularity do not exert a significant impact on expectations10. The intriguing
questions in this context, of course, are: «What is it that influences personal
expectations?»; and «Does the economy at least affect expectations even if it
fails to exert a direct effect on incumbent party popularity?» Table A2 in the
Appendix carries out tests analogous to those shown in Table A1, using
aggregate personal expectations as the dependent variable. The table shows that
the only significant macroeconomic influences on expectations during the 19791994 period were (1) changes in real interest rates (which exert a negative effect:
when real interest rates rise, expectations fall) and (2) the volume of transactions
in the housing market (which exert a positive effect: when transactions are
buoyant, so are expectations). I return to the character and implications of these
various connections below.
Personal economic expectations, however, are by no means the only
economic perceptions that are capable of mediating the relationship between the
objective economy and political support. A second potentially important set of
perceptions concerns attributions of responsibility for economic success or failure.
Individual voters may or may not hold government responsible for whatever is
happening to the economy. If the economic news is good, voters may not credit
the government for it. If the news is bad, they may lay the blame on someone or
something else. In circumstances such as these, we would clearly expect to
observe no relationship between objective economic change and support for the
government. Unfortunately, the problem with the attribution question is that there
are no continuous, over-time data available which might enable us systematically
to measure the extent to which governments are held culpable for economic
success or failure. To be sure, there are some fragmented data which relate to
the position of the current government. (These suggest that there has been a
marked change since the last election in the extent to which voters blame the
government for the length and depth of the 1990-1993 recession. By the spring of
1994, almost three quarters of Gallup’s respondents believed that the government
was either «very» or «somewhat» responsible). The lack of a sustained timeseries, however, means that in the UK context the long-term role of voters’
perceptions of government culpability simply cannot be assessed.
A third set of economic perceptions is potentially the most important of all.
These concern the extent to which the government is seen as being competent to
manage the economy. Survey questions about the relative merits of the Tories
and Labour as economic managers have been asked periodically since 1964.
Almost without exception (one notable example being at the time of the
introduction of the poll tax in March 1990), the Conservatives have been more
favourably regarded than Labour. Not surprisingly, perhaps, the party which has
traditionally been viewed (rightly or wrongly) as representing the interests of
business has tended to be seen as the one most able to run a mixed economy.
Since the beginning of 1991 Gallup have included an economic
competence question in their monthly Gallup 9000 survey. This means that, for
the first time, we now have a continuous series of observations which measure
mass perceptions of the relative economic competence of the two major parties.
Figure 9 shows what happened to these perceptions between January 1991 and
August 1995 -and how changes in them appear to have corresponded to changes
in support for the government. Until the middle of 1992, the competence series is
above the zero line -reflecting the clear lead which the Conservatives had
apparently enjoyed for most of the period since 1964. Note, however, what
happened at the time of Britain’s ignominious exit from the ERM in September
1992. The competence graph (as well as the popularity graph) plunges
downwards and continues to trend downwards thereafter. Although there can be
little doubt that the Conservatives have only themselves to blame for this
downward shift in their fortunes, it is worth emphasising the complementary role
played by the Labour Party. Since the mid 1980s Labour has progressively shed
the image of extremism and irresponsibility in which it cloaked itself as a result of
the 1983 manifesto. The 1989 Policy Review moved Labour back towards the
centre ground of British politics. Since the 1992 election, John Smith’s
Presbyterian rectitude and Tony Blair’s evident moderation have further extended
Labour’s appeal. The shift in the graph described in Figure 9 owes just as much
to Labour’s self-transformation as it does to the Conservatives’ failure to satisfy
the electorate’s economic aspirations.
It is worth noting, finally, that the Conservative lead on economic
competence that was evident until mid 1992 had almost certainly been a crucial
background resource that predisposed many voters to support the Conservatives
electorally -particularly in times of economic uncertainty. From the time of the
ERM crisis in September 1992, that resource seems to have dissolved. Whether
it was the trauma of the crisis itself that provoked the step-shift in perceptions, or
whether the crisis merely triggered a change in perceptions that, as a result of
three years of recession, was «waiting to happen», is impossible to say. In any
event, there can be no doubt that, since September 1992, the Conservatives
have lost a crucial support cushion which they must restore if they are to stand
any chance of re-election in 1996 or 1997.
Where does all this leave us? What the forgoing discussion suggests is
that the connections between macroeconomic changes and UK government
popularity since 1979 have been fundamentally indirect. The effects of the
macroeconomy during the 1980s and early l990s operated mainly through voters’
personal expectations: when real interest rates rose (fell), so did expectations;
and when expectations rose (fell), so did government popularity. Most other
macroeconomic variables (with the exception of taxation and property
transactions) failed to exert either direct or indirect effects on government
popularity. The sort of two-stage model of Conservative support implied by this
analysis is summarised in Figure 10. The parenthetic reference in the figure to
«economic competence» reflects the potential importance of these perceptions as
implied by the evidence presented in Figure 9 above. The reference to «press
coverage» reflects research conducted for the 1979-1987 period -not developed
here because of lack of data availability- which shows that expectations can be
significantly affected by the way in which the national press reports economic
news11.
THREE FORECASTING MODELS FOR 1996-1997
A number of serious difficulties invariably confront the political forecaster.
Since it is not possible to discuss all of them here -and since they are
summarised elsewhere-12 I focus here on a problem that is often paid relatively
scant attention: that of the length of the estimation period which should be used in
order to generate forecasts. The standard econometric solution to this problem is
simple: use the longest contingent time period available. Although this solution is
entirely satisfactory for a model that is thought to be very well-specified, it is less
attractive in circumstances where many of the possible influences upon the
dependent variable in question simply cannot be effectively measured. In politics,
unlike in macroeconomic relationships, there are many intangibles -symbols,
identities, personalities, policies, external threats and challenges- that change in
ways that are not susceptible to formal modelling. In these circumstances, the
average relationship between a given Yt and Xt, over the last thirty years may not
be as good a guide to their likely relationship over the next three years as their
more recent relationship. Indeed, in political forecasting, until we have contrived
more effective ways of operationalising what is currently unmeasurable, the
choice of estimation period must depend upon the analyst’s judgement as to
which past period the medium-term future is most likely to resemble.
The ensuing analysis tests out the implications of using different estimation
periods in order to produce medium-term forecasts of UK government popularity.
Specifically, I consider the consequences of estimating a particular popularity
function against data from two different (but overlapping) time periods and then
using each of the resultant models to forecast popularity in the same post-sample
period. Thus, a popularity function is developed which is estimated (a) for the
period June 1979-December 1993 and (b) for the period January 1991-December
199313. The parameter estimates from (a) and (b) are then employed to generate
two different sets of popularity forecasts for the period January-November 1994.
The discussion focuses on the differences between these two sets of forecasts
and on their accuracy in predicting actual government popularity levels during
1994. A third function is then developed which incorporates additional data on
economic management perceptions -data that, as noted above, are only available
for the period since January 1991.
Table 1 presents a simple forecasting model for the 1979-1993 period14.
The two-stage model, based on the relationships outlined in Figure 10, is derived
using the standard general-to-specific econometric methodology advocated by
Hendry and his associates15. The initial specification for the popularity equation
embraced a large number of lagged and unlagged candidate predictor variables.
These included: (1) measures of the objective economy such as unemployment,
inflation, disposable income, taxation, and (real and nominal) interest rates -as in
Tables A1-A2, all measured as levels and as changes in levels; (2) a measure of
aggregate personal economic expectations; (3) and a series of potentially
important «event dummies» (such as the Falklands War)16. The final
specification retains only those variables that exerted a significant effect on
Conservative popularity. Similarly, the initial specification for the expectations
equation included the same set of objective macroeconomic measures and event
dummies. It also included a measure of the volume of property transactions -in
recognition of the peculiar importance of the housing market during the 1980s in
fuelling UK consumer confidence. The final specification retains only those
variables that exerted a significant effect on expectations.
Several features of Table 1 are worth highlighting. First, only one of the
objective macroeconomic measures -taxation- exerts a direct (and marginally
significant) effect on Conservative popularity over and above the highly significant
effects of personal expectations. The negative sign on the coefficient indicates
that increases taxation -as common sense would suggest- served to reduce
support for the government. As anticipated in the earlier discussion of Figures 16, none of the other objective measures directly affects support. Second, the
Falklands War (in the spring of 1982) and the removal of Margaret Thatcher (in
November 1990) both boosted government popularity -though, as the coefficient
on the lagged dependent variable indicates, these effects discounted quite rapidly
(at the rate of .83 per month). Third, aggregate personal expectations during the
1979-1993 period were significantly affected by two macroeconomic variables:
changes in real interest rates (lagged by 2 months); and the level of property
transactions (lagged by three months). The signs on both of these variables are in
line with theoretical expectations: reductions in real interest rates and a
strengthening property market both served to raise aggregate expectations.
Fourth, the Falklands War and the removal of Thatcher also boosted
expectations, while the introduction of the poll tax (in March 1990) significantly
reduced them. Again, however, the effects discounted relatively quickly -at the
rate of .80 per month. Finally, it is worth stressing that the reported equations all
pass the standard battery of diagnostic tests17.
Table 2 describes an equivalent model to that shown in Table 1 for the
period January 1993 to December 1993. The specification differs from the longerterm model in Table 1 in two minor respects: (1) the property transactions terms
is dropped from the expectations equation because it is non-significant over the
1991-1993 period; and (2) in the popularity equation, a change in taxation
variable (lagged one month) replaces the simple level of taxation index that
appears in Table 1 -on the grounds that this specification, for the shorter time
period, produces a better fit to the data. The general account of the determinants
of government popularity that the results provide, however, is very similar to that
indicated in Table 1. Popularity is influenced positively by aggregate personal
expectations (note the similarity of the expectations coefficients in Tables 1 and
2) and negatively by taxation18. Expectations, in turn, are influenced by changes
in real interest rates -again lagged by two months.
But if these two models both provide parsimonious, theoretically plausible
and statistically significant descriptions of the respective time periods to which
they refer, how well do they predict changes in Conservative popularity after
December 1993? Table 3 shows the relevant results. It displays the popularity
forecasts for 1994 generated by the two models outlined in Tables 1 and 2,
together with the size of the residual associated with each forecast (the difference
between observed and forecast popularity) and a summary measure of the
overall accuracy of the forecasts. What is clear from an inspection of Table 3 is
that neither the 1979-1993 model nor the 1991-1993 were particularly good at
forecasting what was to happen to Conservative popularity during 1994. The
long-term model is particularly weak in the sense that it consistently
overestimates Conservative support. (Its «root mean» score -its average
prediction error- is 3.5). Indeed, the 1979-1993 model produces a negative
residual of increasing magnitude as the year progresses: by November, it
overestimates Conservative support by almost 5 percentage points. The shortterm model provides some predictive improvement in terms of its lower root mean
of 1.2. However, its pattern of underestimation (January-July) and subsequent
overestimation is worryingly systematic: a convincing set of forecasts would
exhibit a far more random pattern.
Although neither set of forecasts makes a compelling case for the
estimation model on which it is based, the results shown in Table 3 lend some
support to the notion that short-term models may be more appropriate than longterm models in the generation of political forecasts. The statistical model
described in Table 1 undoubtedly reflects the very strong relationship that existed
between Conservative popularity and personal economic expectations during the
1980s and early l990s. It is possible, however, that that relationship was
predicated upon the existence of some other, unmeasured, factor which was
relatively constant between 1979 and, say, 1992. If this was indeed the case,
then a forecasting model based largely on a period when the unmeasured factor
was constant might well prove less satisfactory than an equivalent model based
mainly on a period when the unmeasured factor was changing.
The evidence presented in Figure 9 above corresponds exactly to the sort
of «unmeasured factor» referred to here. Such fragmented evidence as there is
suggests that, during the 1980s and early l990s, the Conservatives consistently
enjoyed a reputation for sound economic management skills. Between January
1991 (when economic management competence was first assessed on a regular
monthly basis) and September 1992, this reputation continued. From the time of
the ERM crisis, however, the Conservatives’ management reputation has gone
from bad to worse. It seems plausible to argue in these circumstances that this
change in voters’ competence perceptions may have damaged the linkage
between expectations and government popularity upon which successive
Conservative chancellors were able to rely between 1979 and 1992. I may be
more optimistic about my future economic prospects but, if I believe strongly that
the main opposition party is best able to manage the economy, I may not want to
preserve the status quo that has produced my optimism. Raising my expectations
under these conditions may not increase my inclination to support the incumbent
government.
What is being suggested, then, is that the greater inaccuracy of the longterm forecasting model, in comparison with the short-term forecasting model, may
result from the greater statistical weight (in terms of the numbers of cases being
analysed) accorded in the long-term model to relationships that were predicated
on voters’ perceptions of the Conservatives as superior economic managers.
More seriously, however, both sets of forecasts shown in Table 3 may be
inaccurate because they fail to take account of an unmeasured variable perceptions of managerial competence- which has changed considerably since
the autumn of 1992.
The obvious course of action in these circumstances is to introduce a
measure of management competence perceptions («competence») into the
analysis -though this necessarily restricts any model estimation to the period
since 1991. There are a number of ways in which «competence» could be
incorporated into the sort of model described in Tables 1 and 2 (a) as an extra
independent variable in either the popularity or the expectations equation; (b) as
an interaction term in combination with expectations in the popularity equation;
and/or (c) as an intervening variable between expectations and popularity in an
additional equation. Although the results are not reported here, all of these
possibilities were tested against data for the 1991-1995 period. What is clear
from these results -and they are anticipated by the strong graphical relationship
between Conservative popularity and competence shown in Figure 9- is that the
impact of competence on popularity is so powerful that it drives out the effects of
personal expectations. This does not necessarily mean, however, that there is no
role for expectations whatsoever. It is clearly possible that voters may view the
governing party as displaying greater competence insofar as it improves their own
personal prospects -even if there are many other (unspecifiable) factors that
influence competence perceptions.
Table 4 sets out a simple 3-stage model of Conservative popularity which
hypothesises: (1) that popularity is directly affected by competence perceptions
(positively) and by taxation (negatively); (2) that competence perceptions are
(positively) affected by expectations and, as is evident from the downward stepshift in Figure 9, by the ERM crisis of 199219; and (3) that expectations, as in
Table 2, are (negatively) influenced by real interest rates20. The results reported
in Table 4, estimated for the 1991-1993 period, clearly support these hypotheses.
Expectations continue to influence popularity but, because their effects are
mediated through competence perceptions, the impact of a unit change in
expectations is considerably less in the Table 4 model (b=.21*.11=.02) than in the
Table 2 model (b=.10). The Table 4 model accordingly dampens down the effects
of expectations, suggesting that the prospects for an expectations-lead recovery
in Conservative fortunes before the next general election are rather less than
might supposed on the basis of the results shown in either Table 2 or Table 3.
Table 5 shows the post-sample forecasts and forecast errors (for JanuaryNovember 1994) for the model estimated in Table 4. Comparison between Tables
3 and 5 is instructive. Whereas the forecasts for the simple 2-stage expectations
model in Table 3 produced marked and systematic forecasting errors, the 3-stage
expectations-competence model in Table 5 (produces much smaller and nonsystematic errors with a noticeably smaller root mean (0.95 in Table 5 as
opposed to 1.21 and 3.50 in Table 3). Put simply, the 3-stage model was far more
accurate in predicting what happened in 1994 than either the long-term or the
short-term 2-stage model.
GENERATING FORECASTS FOR 1996-1997
What does all of this imply for making forecasts about future movements in
UK government popularity? Two broad lessons can, I think, be learned. The first
is that political forecasts based on a shorter, relatively recent estimation period
are more likely to be accurate than forecasts based on a longer and therefore
partly more distant period. In the present context, of course, if competence
perceptions are to be included in the forecasting model, short-term estimation is
the only option anyway because no long-term data exist. Even without this
constraint, however, there is still a strong case for restricting the estimation period
to the relatively recent past. Although the evidence is barely more than
impressionistic, there is an increasing sense among British political and economic
commentators that the very success of Thatcherism in the 1980s, in exposing
many areas of British economic life to genuine market competition, lead directly to
a new climate of economic insecurity in the l990s. Table 6 summarises the
evidence from one of the few systematic attempts to measure the extent of this
new insecurity -which appears to be very considerable indeed. The problem with
such data, of course, is that, without some earlier point of reference, it is
impossible to establish whether the levels of insecurity recorded are increasing,
constant or even decreasing. If we assume, however, that the «new economic
insecurity» thesis does have some validity, then it implies a new set of
background conditions which can clearly be held constant by restricting the
estimation period to the recent past.
The second main implication of the forgoing discussion concerns the
difficulty of deciding which estimation model should be used to forecast
movements in Conservative popularity over the next two years. On the face of it,
the fact that the 3-stage model produced more accurate post-sample forecasts for
1994 than the 2-stage model could be taken to indicate that the 3-stage model is
likely to perform better in 1995-1997. Unfortunately, the choice of models cannot
be made quite so easily -and for at least three reasons.
First, it needs to be recognised that each of the forecast popularity scores
shown in Tables 3 and 5 has a standard error of roughly 3 percentage points
associated with it. Strictly speaking, therefore, each point estimate forecast has a
(two standard error) confidence interval of six percentage points either side of it. It
follows that all of the forecasts reported in Tables 3 and 5 are within the stipulated
forecasting range and that, on a purely statistical basis, we cannot conclude that
one model is preferable to the other. This said, the greater predictive accuracy of
the 3-stage model is bound to increase its credibility as a forecasting tool.
Second, a significant potential weakness of the 3-stage model is that it
encompasses a statistical series -the competence index- that has relatively few
major «turning points». The series has certainly been subjected to a marked
downward shift (in October 1992) and to a partial recovery (in December 1992).
However, as the results in Table 2 indicate, these turning points are modelled as
«one-off» dummy variables and accordingly there is no guarantee that the
remaining predictor variable in the competence equation -personal expectationswill be capable of modelling any future turning point(s) in the series.
Third, the competence equation in the 3-stage model is neither statistically
robust nor particularly illuminating from a substantive viewpoint. Although, as
Table 4 shows, expectations exert a significant effect on competence during the
1991-1993 period, as shown below this effect weakened considerably during the
course of 1994. The reason for this weakening is shown clearly in Figure 11: the
partial recovery in expectations during the summer and autumn of 1994 was not
accompanied by an improvement in the Conservatives’ competence ratings.
When this weak association between expectations and competence is combined
with the failure of any other macroeconomic variables to exert any sort of effect
on competence, the substantive limitations of the competence equation shown in
Table 4 become very clear indeed. Competence is clearly the major influence on
popularity, but competence itself is extraordinarily difficult to explain (and
therefore to forecast) statistically.
In these circumstances -where the 2-stage model is not particularly good
at forecasting the post-sample period and the 3-stage model has both statistical
and substantive limitations- it is tempting to abandon the attempt to forecast the
next UK general election altogether. Such a course of action, however, would be
unnecessarily timid. A forecasting model is a statement of the analyst’s beliefs
about the nature of the processes that are operating during her/his estimation
period. I am prepared to commit myself to the proposition that something like
either the 2-stage or the 3-stage model has operated in the UK during the 19911994 period. I am simply unable to determine which is the more plausible. I
therefore provide two alternative sets of forecasts -one corresponding to each
model. I leave it to the reader, and to future observation, to determine which, if
either, is more appropriate.
The forecasting equations that I employ are summarised in Table 7 and 8.
The equation specifications are identical to those presented in the relevant parts
of Tables 2 and 4. The estimation is conducted for a longer sample period
(January 1991-November 1995). Although there are minor variations, the
reported coefficients in Tables 7 and 8 are similar to the equivalent ones shown in
Tables 2 and 4.
In order to produce forecasts of Conservative popularity, of course, it is
first necessary to make assumptions about changes in the exogenous variables
specified in the models. In both the 2-stage and 3-stage models, there are only
two wholly exogenous variables: changes in real interest rates and changes in
taxation. It should be recalled that other candidate predictors were included in the
initial model specifications but that no other independent variables yielded
significant coefficients. There is no reason to suppose, therefore, that changes in
other macroeconomic variables would affect any of the endogenous variables in
either of the models. This absence of other predictor variables is a measure of the
present Conservative government’s lack of room for politico-economic
manoeuvre. The only effective macroeconomic levers that it can pull in order to
maximise its own prospects of re-election are to attempt to reduce either real
interest rates or taxes -or both.
Economic and political realities, however, are likely to prevent the
Chancellor from cutting either real interest rates or taxes to the extent that would
be necessary to achieve a sustained recovery in the Conservative party’s poll
ratings. Real interest rates in the year to August 1995 were approximately 4%.
Moreover, given the likely reaction of the currency markets, however, progressive
real rate reductions in excess of 2 percentage points would appear to be
financially impossible. Thus, the government’s interest rate lever -which it pulled
so effectively in the run-ups to the 1983, 1987 and 1992 general elections, is
severely constrained. There is perhaps a little more leeway in terms of possible
tax reductions. The government was obliged to increase taxation significantly in
1993 and 1994. It clearly hopes that public finances will be sufficiently robust by
1996 that significant reductions in taxation can be introduced. Massive tax
reductions immediately before an election, of course, risk being interpreted by
voters as electoral bribes that (as in 1992?) would probably be withdrawn almost
as soon as the election had been won. It is therefore important, from a political
standpoint, to introduce any tax reductions gradually several months before it is
intended that an election is to be called.
With these constraints in mind, I make the following assumptions about
real interest rate and taxation reductions over the next 2 years.
(1) The government has a target date for the next election already in mind.
Given the government’s current low standing in the polls, I assume that it will
want to delay an election as long as possible. (At the same time, however, if the
government waits until the last possible moment -May 1997- it risks losing all
freedom of manoeuvre if an unforeseen crisis arises). Nonetheless, I assume that
the government’s current target date for holding the next election is April 1997.
(2) The government recognises that the effects of any real interest rate or
tax reductions that it makes will take time to feed through to voters’ political
preferences. The lags and discount rates in the various models specified in Table
7 suggest that the full effects of these changes take at least 3 months to work
through.
(3) The government recognises that any interest rate or tax reductions
have to be made gradually over a period of at least 6-9 months prior to the target
election date in order to persuade voters that any resultant improvement in their
economic perceptions is relatively durable. (Massive tax and/or real interest rate
reductions in the month or two before an election could not be presented as
anything other than the most crass political manipulation and would almost
certainly be counterproductive).
(4) In accordance with the logic of assumptions (1)-(3), the Chancellor
initiates a surreptitious policy of progressive real interest rate reductions from the
beginning of 1996. Either by reducing nominal rates or by allowing inflation to rise
while nominal rates are held constant, real rates are reduced by two percentage
points between January and December 1996.
(5a) In accordance with the logic of assumptions (1)-(3), the Chancellor
initiates a surreptitious series of tax reductions in November 1995 which follow a
similar pattern as the tax reductions that Chancellor Lamont introduced in the 18
months before the 1992 election. In terms of the taxation index employed here,
this involves reducing taxation to a score of -1 by January 1997.
(5b) In accordance with the logic of assumptions (1)-(3), the Chancellor
initiates a surreptitious series of tax reductions as in (5a) above, but between
November 1995 and January 1997 s/he introduces a sufficiently large tax cut to
reduce the taxation index to the lowest level achieved under Nigel Lawson (in
April 1988) when the taxation index stood at -2.2.
Table 9 reports the simulated results of applying these assumptions to the
two forecasting models shown in Tables 7 and 8. The simulations, if they have
any validity at all, provide depressing news for the Conservatives. If the 2-stage
model is to be believed, even on the most generous assumptions about tax and
interest rate reductions (that is, assumptions 4 and 5b) the Conservatives’
forecast popularity rises no higher than 36%. This is over 7 percentage points
short of the 43% target needed to ensure re-election. If the more cautious 3-stage
model is employed, however, forecast popularity rises to just above 31% on
assumption 5b and reaches under 30% on assumption 5a. Such levels of
support, if they were actually to be attained, would in all probability leave the
Conservatives with less than 150 seats in the next Parliament.
A final set of simulated results is provided in Table 10. The figures reported
assume that, regardless of what happens to real interest rates, the government is
able (by whatever devices) to raise the level of aggregate personal expectations
to a score of +5 -an expectations level equivalent to that achieved in June
198321. On this assumption the 3-stage model predicts a Conservative vote
share of 33% if taxation can be reduced to 1992 levels and of 34.2% if taxation
can be reduced to 1988 levels. The 2-stage model, on the other hand, predicts
39% under 1992 taxation assumptions and fully 41% if taxation can be brought
down to the levels voters enjoyed in 1988. Before Conservative supporters clutch
at these latter two statistical straws, however, it should be emphasised (a) that
there is no evidence to suggest that expectations will rise as high as +5 points in
the run-up to 1997; and (b) that the expectations-government popularity
relationship may itself have been fractured by the failure of the Conservatives, in
the wake of the ERM crisis, to sustain their reputation for competent economic
management.
Moreover, all of these simulated results need to have two very prominent
caveats attached to them. First, the macroeconomic assumptions underlying the
simulated forecasts may themselves turn out to be inaccurate -in which case, the
forecast figures would not be directly relevant anyway. Second, as noted earlier,
each of the forecasts has a confidence interval of six percentage points either
side of it. In strict statistical terms, the simulated result of 36% referred to above
says that the Conservative share of the vote will be between 30% and 42%
(Some prediction!). Although it counsels caution, this technical limitation does not
mean that the forecasting exercise is valueless. While it is true that we can only
be 95% confident, given the specified assumptions, that a popularity forecast of,
say, 32% will be associated with an observed popularity score somewhere
between 26% and 38%, this is not the only interpretation that can be made of the
forecast. Another -and equally valid- interpretation is to regard the 32% «point»
forecast as the best guess that we can make, on the basis of the available
empirical evidence, of the consequences that certain specified macroeconomic
changes are likely to have for Conservative popularity. Of all the possible
consequences of the specified exogenous changes, the point forecast represents
the one most likely to occur. The confidence interval represents the size of the
«health warning» -and, in the present context, it is clearly a considerable onethat needs to accompany any forecast. The forecasts developed here are not
presented as firm predictions of what will happen in the spring of 1997. Rather,
they are presented -extremely tentatively- as a loose indication of the sort of
recovery in support that the Conservatives can reasonably expect to achieve as
their re-election strategy develops over the next year or so.
SUMMARY AND CONCLUSIONS
The condition of the domestic economy exerts a profound influence on the
political preferences of British voters -but it does so in a variety of complex and
changing ways. During the 1980s and early l990s, with (as far as we can tell) a
clear majority of voters believing in the superior economic management skills of
the Conservatives, successive Chancellors were able to pursue macroeconomic
strategies -based around real interest rate and taxation reductions- designed to
raise voters’ economic expectations in advance of each general election. In 1983,
1987 and 1992, this raising of expectations duly assisted the government in
securing re-election.
The relatively stable connections between support for the government and
aggregate personal expectations over the 1979-1992 period, moreover, was also
rather useful to political scientists. It meant that medium-range political forecasts
could be made with some degree of confidence. The government could raise
expectations by manipulating real interest rates and taxation; it would almost
certainly attempt to do so; and if it made the right sort of reductions in both, it
could expect to increase its support by a specified amount.
Due to a complex set of factors which are probably impossible to model
given the sort of time-series data currently available, it is possible that the
expectations-government popularity relationship of the 1980s has weakened
significantly in the mid l990s. Voters’ economic perceptions continue to be central
to the political fortunes of the government. Since 1992, however, perceptions of
economic management competence appear to have mattered more than
expectations. It is in this context that both the present Conservative government
and political forecasters are encountering difficulties. The Conservatives appear
to have lost their reputation for competence as a result of a combination of factors
-the length and depth of the 1990-1993 recession; the export-lead recovery which
has been unaccompanied by a revival in domestic consumer confidence; and the
new climate of post-Thatcherite economic insecurity- which were crystalised in
voters’ minds at the time of the ERM crisis in September 1992. The government
apparently has little sense as to how it can revive its previous reputation for
managerial competence. And it is here that the problems of the government and
the forecaster coalesce. It is very difficult to devise a good statistical model which
explains why that reputation was so dramatically lost in 1992. As a result, the
forecaster has no compelling basis for making even contingent forecasts about
future movements in the government’s competence ratings.
This said, I am confident that, if the Conservatives’ competence ratings
rise dramatically, the party will enjoy a corresponding recovery in its support. This
reflects my conviction that the explanation for the Conservatives’ current electoral
difficulties lies primarily in the loss of their previous reputation for sound economic
management. Unfortunately, it is extremely difficult to specify the conditions under
which the government’s reputation for management competence might be
restored.
In these circumstances, the most plausible forecasts that can be made are
probably those which rely on the continuing operation of the (2-stage)
expectations-popularity relationship. Given this assumption, the simulations
conducted here suggest that, if the government is able to survive the projected
loss of its Commons majority during 1996, then purely on the basis of economic
considerations the Conservatives can expect to achieve a popularity rating of
around 36% by the spring of 1997. Although this figure is almost 10 percentage
points higher than the level currently recorded in the opinion polls, it would clearly
not be sufficient to produce a Conservative victory in the next general election.
Table1
Table2
Table3
Table4
Table5
Table6
Table7
Table8
Table9
Table10
TableA1
TableA2
Figure1
Figure2
Figure3
Figure4
Figure5
Figure6
Figure7
Figure8
Figure9
Figure10
Figure11
NOTES
The paper discusses the prospects for successfully forecasting the outcome of the next UK General
election. It focuses on the problem of deciding which estimation period should be used in order to produce a
particular set of political forecasts. A series of forecasting equations are developed based partly on different
estimation periods and partly on different assumptions about which particular economic perceptions will prove
decisive in British politics over the next year or so. Although the precise results of the various forecasting
models differ, they all converge on the conclusion that the Conservatives are unlikely to recover sufficiently on
the basis of purely economic considerations to win the next general election.
1. The seminal study in the UK was GOODHART, C.A.E. and BHANSALI, R.J.: «Political Economy»,
Political Studies, vol. 18/1970, p. 43-106.
2. See, for example, WILLIAM, Miller and MACKIE, W.M.: «The Electoral Cycle and the Asymmetry of
Government and Opposition Popularity», Political Studies, vol. 21/ 1973, p. 263-279; PISSARIDES,
Christopher: «British Government Popularity and Economic Performance», Economic Journal vol. 90/ 1980, p.
569-581; BOROOAH, Vani K. and VAN DER PLOEG, Frederic: «British Government Popularity and
Economic Performance: A Comment», Economic Journal, vol. 92/ 1982, p. 405-410; CLARKE, Harold,
STEWART, Marianne C. and ZUK, Gary: «Politics, Economics and Party Popularity in Britain, 1979-1983»,
Electoral Studies vol. 5/1986, p. 123-141; CLARKE, Harold, MISHLER, William and WHITELEY, Paul:
«Recapturing the Falklands: Models of Conservative Popularity, 1979-1983», British Journal of Political
Science, vol 20/ 1991, p. 63-82; DUNLEAVY, Patrick and HUSBANDS, Christopher T.: British Democracy at
the Crossroads: Voting and Party Competition in the 1980s. London, George Allen and Unwin, 1985; FREY,
B. and SCHNEIDER, F.: «A Politico-Economic Model of the United Kingdom», Economic Journal, vol. 88/
1978, p. 375-394; ALT, James: The Politics of Economic Decline: Economic Management and Political
Behavior in Britain Since l964. Cambridge, Cambridge University Press, 1979; NORPOTH, Helmut: «Guns
and Butter and Economic Popularity in Britain», American Political Science Review vol. 81/ 1987, p. 949-959;
NORPOTH, Helmut: «The Popularity of the Thatcher Government: A Matter of War and Economy» in
NORPOTH, Helmut, LEWIS-BECK, Michael S. and LAFAY, Jean-Dominique (eds.): Economics and Politics:
The Calculus of Support. Ann Arbour, University of Michigan Press, 1992; PRICE, Simon and SANDERS,
David: «Modeling Government Popularity in Postwar Britain: A Methodological Example» American Journal of
Political Science, vol. 37/ 1993, p. 317-334.
3. SANDERS, David, MARSH, David and WARD, Hugh: «Government Popularity and the Falklands War»,
British Journal of Political Science vol. 17/ 1987, p. 281-313; SANDERS, David, MARSH, David and WARD,
Hugh: «A Reply to Clarke, Mishler and Whiteley», British Journal of Political Science, vol. 20/ 1990, p. 83-90;
SANDERS, David, MARSH, David and WARD, Hugh: «Macroeconomics, the Falklands War and the
Popularity of the Thatcher Government: A Contrary View» in NORPOTH, LEWIS-BECK and LAFAY (eds.):
Economics and Politics, op. cit., p. 161-184.
4. SANDERS, David: «Government Popularity and the Next General Election», Political Quarterly, vol. 62/
1991, p. 235-261. CLARKE, Harold, STEWART, Marianne C. and WHITELEY, Paul: «Tory Trends: Party
Identification and the Dynamics of Conservative Support Since 1992», British Journal of Political Science,
forthcoming 1997. It should be noted that Clarke et al’s recent work examines the importance of party
identifications and leadership factors, as well as economic perceptions, in the determination of party support
patterns. I eschew the use of these additional predictors here on the grounds that it is almost impossible at
present to generate independent forecasting equations for either identification levels or leadership ratings. If a
party’s identification ratings increase, or if its leader’s ratings rise, then it is hardly surprising if its opinion poll
standing improves. The difficult forecasting task is to determine why identification or leadership ratings have
improved in the first place: and we currently have no basis for forecasting either of these phenomena.
5. SANDERS, David: «Forecasting the 1992 General Election Result: The Performance of an Economic
Model» in DENVER, David et al.: British Elections and Parties Yearbook 1993. London, Harvester
Wheatsheaf, 1993, p. 100-117.
6. SANDERS, David: «Economic Influences on the Vote: Modelling Electoral Decisions» in BUDGE, Ian and
McKAY, David (eds.): Developing Democracy: Research in Honour of J.P.F. Blondel. London, Sage, 1992, p.
79-97.
7. The taxation index here is derived from the CSO’s tax and price index. It is obtained by subtracting the
CSO’s price index from the tax and price index.
8. Figure 8 shows the connections between government popularity and aggregate personal prospections.
Over the 1979-1993 period, aggregate neasures of (i) personal retrospections, (ii) general prospections and
(iii) general retrospections all correlate with government popularity to varying degrees (respectively, r = .51,
.33 and .45) but none correlates with popularity as strongly personal prospections (r = .72).
9. Standard recursive coeficient tests and rolling recursive coefficient tests show that the value of the
expectations coefficient remains stable throughout the 1979-1994 period.
10. Specifically, over the period June 1979 to November 1994:
CONt= 4.66 + .88 CONt-1 + .06 PEXPT-1 + ut
(1.40) (.03)
(.02)
CONt= 4.52 + .88 CONt-1 + .05 PEXPt-2
+ ut
(1.38) (.03)
(.02)
PEXPt= -7.71 + .80 PEXPt-1 + .16 CONt-1 + ut
(3.52) (.05)
(.09)
PEXPt= -3.92 + .85 PEXPt-1 + .07 CONt-2 + ut
(3.39)
(.05)
(.08)
were CON is (monthly) Conservative Popularity, PEXP is aggegate personal economic expectations and ut is
random error term. Standard errors in parentheses.
11. SANDERS, David, WARD, Hugh and MARSH, David: «The Electoral Impact of Press Coverage of the UK
Economy» British Journal of Political Science, vol 23/ 1993, p. 175-210.
12. SANDERS, David: «Forecasting Political Preferences and Election Outcomes in the UK: Experiences,
Problems and Prospects for the Next UK General Election» Electoral Studies, vol 14/ 1995, p. 251-272.
13. The June 1979 date is chosen because represents the beginning of an extended period of Conservative
government. The January 1991 date is chosen because this represents the beginning of a period for which
additional economic perceptions data (employed in a subsequent model) are available.
14. In the context of this paper, I do not enter the protracted debate as to whether timeseries data should
invariably be de-trended and how popularity functions should be specified and estimated. I simply note that all
of the timeseries employed here can be de-trended by first-differencing; that they can therefore be regarded
as a co-integrating set; and that the lagged endogenous variable model which embraces either levels or
changes variables is accordingly appropriate for estimation purposes. For general discussion of these issues,
see SANDERS, David and WARD, Hugh: «Timeseries Techniques for Repeated Cross-Section Data» in
DALE, Angela and DAVIS, Richard B. (eds.): Analysing Social and Political Change: A Casebook of Methods.
London, Sage, 1994, p. 198-223.
15. HENDRY, David: «Economics-Alchemy or Science», Economica, vol. 46/1980, p. 387-406; HENDRY,
David: «Econometric Modelling: The ‘Consumption Function’ in Retrospect», Scottish Journal of Political
Economy, vol. 30/ 1983, p. 193-220.
16. The effects of «unusual events» are, of course, relatively easy to model after they have occurred. It is far
more difficult to forecast such effects in advance. Generally, however, the effects on party support of most
«events» discount quite rapidly. (With the specifications employed here, the decay rate is defined by the
magnitude of the coefficient on the lagged dependent variable). All of the forecasts provided here assume that
Conservative support will be unaffected by «unusual events» in the period through to the next general
election.
17. It is important to note in this context that Dickey-Fuller (DF) tests for possible unit roots in the error term
are applied to each of the OLS models tested in this paper. (For the rationale underlying these tests, see
HARRIS, Richard: Using Cointegration Analysis in Econometric Modelling. London, Prentice Hall-Harvester
Wheatsheaf, 1995). All of the reported DF tests are above the required critical value, indicating that in each
case the null hypothesis of non-stationarity can be rejected -i.e. that the residuals in each equation are indeed
stationary. Given that the residuals also pass all the other standard diagnostic tests relating to serial
correlation, functional form and heteroscedasticity, it is assumed throughout that there is no need to apply the
principles of co-integration theory by using error correction models. On the rare occasions where the OLS
specification produces residuals do not pass one of the standard tests, maximum likelihood estimation is
employed. This strategy of (a) eschewing error correction models and (b) deploying maximum likelihood
methods when OLS assumptions are violated means that the models reported here do not seek to «recover»
any long-term equilibrium effects. In the context of the forecasts that are being made here, this limitation is not
serious -and for two reasons. First, given the shortness of the forecasting period, it is entirely reasonable to
place the main emphasis on specifying and estimating the short-term dynamics of the various relationships
involved. Second, it is quite possible that there is no long-term equilibrium (of the sort posited in error
correction specifications) between the sorts of attitudinal measures that form the core of the models that are
estimated here. This conclusion is certainly supported by the evidence reported below which shows that short-
term estimation models provide more accurate short-term forecasts than their long-term counterparts suggesting that the relationships among the core attitudinal variables vary over time.
18. The difference in coefficient magnitudes between Tables 1 and 2 in this context reflects the different
metrics of the levels and change in taxation variables.
19. The impact of the September 1992 ERM crisis was initially modelled separately using an extensive series
of dummy variables. It was clear from this modelling exercise that the initial effect of the crisis was to reduce
competence perceptions by roughly 20 percentage points. Within two months, this shock had dissipated
somewhat, with competence rising in December by around 10 percentage points. Thereafter, the competence
ratings stabilised -some 10 points lower than they had been prior to the crisis. The ERM terms in the equation
in Table 4 accordingly refer to dummies on October and December 1992.
20. The reported equations are derived using the same Hendry general-to-specific methodology employed for
Table 1, with the same set of candidate predictors (unemployment, inflation, disposable income, interest rates,
taxation) being included in the initial model specification for each equation.
21. This assumption may not be implausible. During the course of 1996, the first generation of TESSA savings
plans are set to mature; at least one merger between mayor building societies is planned; and one prominent
society may become a bank. All of these developments can be expected to raise their beneficiaries’
disposable incomes and thereby to raise the overall level of aggregate personal expectations in unpredictable
ways -and independently of any changes in expectations that derive downward movements in real interest
rates.